I am Linfeng Zhao (赵林风), a CS Ph.D. student at Khoury College of Computer Sciences of Northeastern University. I’m advised by Prof. Lawson L.S. Wong and also working with Prof. Robin Walters. During PhD: I am collaborating with and visiting MIT LIS group led by Prof. Leslie Kaelbling and Prof. Tomás Lozano-Pérez. I interned at Meta (2024 summer), Boston Dynamics AI Institute (2023 Spring-Summer, WUD team with Jennifer L. Barry) and earlier at Amazon Science (2021 Summer). During undergraduate: I interned at Microsoft Research Asia (2019) and earlier worked with Prof. Hao Su at UC San Diego (2018-19).
I focus on decision-making in robotics and work on the intersection of machine learning, robotics and artificial intelligence. My current research centers on improving open-world and long-horizon decision-making for mobile-manipulation robots, like Boston Dynamics Spot, which involves planning (in task and motion level) and learning (state/action representations, world models, skill policies). My recent work involves improving scalability, generalizability, and efficiency for robotic decision-making, including utilizing knowledge and capabilities in pre-trained multimodal models and leveraging symmetry and compositional structures.
We study the problem of action sampling and propose a method to incorporate equivariance properties to the action sampling procedure.
This work learns to navigate end-to-end with map input, aiming to generalize to novel map layouts in zero-shot.
We enable a robot to rapidly and autonomously specialize parameterized skills by planning to practice them. The robot decides what skills to practice and how to practice them. The robot is left alone for hours, repeatedly practicing and improving.
We study E(2) Euclidean equivariance in navigation on geometric graphs and develop message passing network to solve it.
We study whether Euclidean symmetry can help in reinforcement learning and planning, which models the geometric transformations between reference frames of robots.
We formulate how differentiable planning algorithms can exploit inherent symmetry in path planning problems, named SymPlan, and propose practical algorithms.
We study how implicit differentiation helps scale up and improve convergence of differentiable planning algorithms.
PDF Poster Slides Program Page (WAFR 2024) arXiv (Coming soon) Code (Coming soon)
PDF Poster Slides Program Page (RLC 2024) arXiv (Coming soon) Code (Coming soon)
PDF Code Poster Slides ICLR page OpenReview arXiv Webpage (Available soon)
PDF Code Poster Slides ICLR page OpenReview arXiv Webpage (Available soon)
PDF Code ICML Page Website (Available soon) Poster (ICML) Slides (ICML oral)